从string解析graph_def时出错

Question

我试图运行一个非常简单的Tensorflow图保存为.pb文件,但在解析它时我有这个错误:

Traceback (most recent call last):
  File "test_import_stripped_bm.py", line 28, in <module>
    graph_def.ParseFromString(fileContent)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/message.py", line 185, in ParseFromString
    self.MergeFromString(serialized)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/python_message.py", line 1069, in MergeFromString
    if self._InternalParse(serialized, 0, length) != length:
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/python_message.py", line 1105, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 633, in DecodeField
    if value._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/python_message.py", line 1105, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 612, in DecodeRepeatedField
    if value.add()._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/python_message.py", line 1105, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 743, in DecodeMap
    if submsg._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/python_message.py", line 1095, in InternalParse
    new_pos = local_SkipField(buffer, new_pos, end, tag_bytes)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 850, in SkipField
    return WIRETYPE_TO_SKIPPER[wire_type](buffer, pos, end)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 799, in _SkipGroup
    new_pos = SkipField(buffer, pos, end, tag_bytes)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 850, in SkipField
    return WIRETYPE_TO_SKIPPER[wire_type](buffer, pos, end)
  File "/usr/local/lib/python3.5/dist-packages/google/protobuf/internal/decoder.py", line 814, in _SkipFixed32
    raise _DecodeError('Truncated message.')
google.protobuf.message.DecodeError: Truncated message.

这是我用来将其写入.pb的代码:

import tensorflow as tf

builder = tf.saved_model.builder.SavedModelBuilder('models/TEST-3')

w1 = tf.Variable(tf.random_normal((2,2)), name="w1")
w2 = tf.Variable(tf.random_normal((2,2)), name="w2")

sess = tf.Session()
sess.run(tf.global_variables_initializer())

builder.add_meta_graph_and_variables(sess, tags=[tf.saved_model.tag_constants.SERVING], clear_devices = True)

builder.save()
sess.close()

这是解析它的代码:

import tensorflow as tf
import os

model_path = os.path.join('models/TEST-3', 'saved_model.pb')
with open(model_path, mode='rb') as f:
    fileContent = f.read()
graph_def = tf.GraphDef()
graph_def.ParseFromString(fileContent)

要查看我必须要做的确切错误

export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python

在运行它之前.另外我在python 2和3上尝试了这个,使用不同的tensorflow版本,我在Ubuntu 16.04上运行.在具有tensorflow 0.9.0rc0的python 2.7上,我设法得到了一个稍微不同的错误:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/message.py", line 185, in ParseFromString
    self.MergeFromString(serialized)
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/python_message.py", line 1091, in MergeFromString
    if self._InternalParse(serialized, 0, length) != length:
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/python_message.py", line 1127, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/decoder.py", line 633, in DecodeField
    if value._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/python_message.py", line 1127, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/decoder.py", line 612, in DecodeRepeatedField
    if value.add()._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/python_message.py", line 1127, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/decoder.py", line 612, in DecodeRepeatedField
    if value.add()._InternalParse(buffer, pos, new_pos) != new_pos:
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/python_message.py", line 1127, in InternalParse
    pos = field_decoder(buffer, new_pos, end, self, field_dict)
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/decoder.py", line 489, in DecodeRepeatedField
    value.append(_ConvertToUnicode(buffer[pos:new_pos]))
  File "/usr/local/lib/python2.7/dist-packages/google/protobuf/internal/decoder.py", line 469, in _ConvertToUnicode
    return local_unicode(byte_str, 'utf-8')
UnicodeDecodeError: 'utf8' codec can't decode byte 0x80 in position 18: 'utf8' codec can't decode byte 0x80 in position 18: invalid start byte in field: tensorflow.FunctionDef.Node.ret

我可以使用此代码解析其他.pb图,例如这一个https://github.com/taey16/tf/blob/master/imagenet/classify_image_graph_def.pb

提前致谢.

Answer 1

这里的问题是你试图解析一个SavedModel协议缓冲区,好像它是一个协议缓冲区GraphDef.虽然SavedModel包含GraphDef,但它们具有不同的二进制格式.以下代码,使用tf.saved_model.loader.load()应该工作:

import tensorflow as tf

with tf.Session(graph=tf.Graph()) as sess:
    tf.saved_model.loader.load(
        sess, [tf.saved_model.tag_constants.SERVING], "models/TEST-3")